Abnormal Data Generation for Deep Learning Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence technologies face challenges in updating deep learning models when data changes, particularly in processing complex non-linear or dynamic data sets, which affects the accuracy and efficiency of data processing in industrial applications.
Innovation Solution
A method involving a computer program executed by processors that selects and transforms data within feature groups, assigns labels, and calculates transformed data using a model, including clustering data subsets and using a cost function based on triplet loss for classification, to improve data processing and model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning models are used to process complex non-linear or dynamic data, then the ability to learn complex patterns is improved, but the difficulty of updating models when data changes increases
Solution Approach 1:
The patent segments the model updating process into multiple stages: generating abnormal data samples through data transformation, training the model incrementally with these samples, and updating only specific components of the deep learning model rather than retraining the entire model. This segmentation reduces the complexity of model updates while maintaining the ability to learn complex patterns.
Solution Approach 2:
The patent performs preliminary actions by pre-generating abnormal data samples and preparing training data before actual model updates are needed. This preliminary preparation includes transforming normal data into abnormal data samples and organizing them for incremental training, which simplifies the subsequent model updating process when data changes occur.
2Measurement precision
If separate abnormal data construction is performed for training and testing, then the accuracy of model performance evaluation is improved, but the complexity of the training and testing processes increases
Solution Approach 1:
The patent merges the abnormal data construction process into the existing data transformation pipeline. By integrating abnormal data generation with normal data transformation operations and using the same data processing infrastructure for both training and testing, the system maintains high measurement precision while reducing process complexity.
Solution Approach 2:
The patent creates a universal data transformation system that can generate both training data and testing data through the same abnormal data generation mechanism. This multi-functional approach allows the system to evaluate model performance accurately while avoiding the need for separate, complex data construction processes for training and testing.
Data Source
AI summary
In an exemplary embodiment of the present disclosure, disclosed is a computer program stored in a computer readable storage medium executable by one or more processors, in which when the computer program is executed by one or more processors of a computing device, the computer program performs operations below for processing data, the operations may include: selecting a plurality of different data from a data set including data formed of one or more feature groups, transforming a part of each data among the plurality of selected data, assigning a label to each of the plurality of transformed data, and computing the transformed data by using the model.